feat: add omni-scient-0.1.0 champion-challenger ensemble

[suryanshsrivastava]

Omni-Scient v0.1.0 — Champion-Challenger Ensemble

A deterministic, pure-stdlib forecasting ensemble. Part of the Omni model family.

Architecture

• Segmentation: Classifies each series as STABLE, SEASONAL, LUMPY, VOLATILE, or NEW
• Candidates: SNAIVE (seasonal naive), AVG3 (trailing average), SIDX (seasonal index × trend), Holt-Winters additive
• Champion selection: Per-series WAPE on a held-out validation window
• Forecast: Champion retrained on full history

Characteristics

• Zero pretrained dependencies — trains from scratch on each task
• Pure stdlib — only numpy beyond the Python standard library
• Deterministic — same input always produces same output
• Includes quantile estimates via residual std dev scaling
• Graceful NaN/inf handling

Files

• models/omni-scient/model.py — fev model wrapper
• models/omni-scient/requirements.txt — pinned deps
• benchmarks/fev_bench/results/omni-scient-0.1.0.csv — full benchmark (100/100 tasks)

Vision

omni-scient is being developed as a foundational time series forecasting model.
This v0.1 is a narrow champion-challenger baseline — the first step toward a pretrained
foundation model in the Omni family.

[shchur]

Hi @suryanshsrivastava thanks for your contribution. The scores for the proposed model look much worse than currently the worst baseline considered in the benchmark. I don't think we want to include the proposed model to the leaderboard at this point.

model_name	win_rate	skill_score
Chronos-2	84.6	47.3
TiRex-2	81.6	45.5
Toto-2.0-2.5B	81.1	44.4
Toto-2.0-1B	80.4	44.4
Toto-2.0-313m	78.6	44.2
TiRex	74.6	42.6
Toto-2.0-22m	72.5	42.9
TimesFM-2.5	71.1	42.2
TabPFN-TS-3	66.9	43.1
FlowState	65.2	39.9
Toto-1.0	62.8	40.8
Toto-2.0-4m	61.8	40.7
TabPFN-TS	61.7	41.5
Moirai-2.0	58	39.3
Chronos-Bolt	57.5	38.9
TFT	43.9	32.2
Sundial-Base	40.1	33.4
Stat. Ensemble	39.2	20.7
PatchTST	38.5	30.1
DeepAR	37.7	29.1
AutoARIMA	34.7	21.3
CatBoost	29.3	23
AutoETS	29.2	-26.8
LightGBM	27.5	21
AutoTheta	24.5	5.5
Seasonal Naive	17.2	0
Naive	11.3	-45.4
Drift	9.9	-45.8
omni-scient-0.1.0	8.4	-91.7